Analyze data and spreadsheets

Got a spreadsheet, an export, or a CSV you’d rather not read row by row? Attach it to a conversation and ask Chat to do the reading for you — totals, trends, outliers, breakdowns by category, the rows that don’t add up. Chat reads your data, answers in plain language, and lays results out as clean tables you can copy straight into your own work. This guide takes you from a raw file to an analysis you can act on.

What you’ll be able to do

Pull a spreadsheet, CSV, or other data file into a conversation.
Ask for totals, breakdowns, trends, comparisons, and summaries — in everyday words.
Get results back as readable tables you can copy or export.
Save a useful analysis and re-run it on next period’s file.

Before you start

You’re in a conversation (a new chat, a folder, a channel, or a direct message).
Your data file is one of the supported formats and under the 100 MB size limit.
Your data is laid out cleanly: a header row of column names, one table per sheet, one value per cell. Tidy data gives far better answers.
The data is something you’re comfortable analyzing in Chat — see Keep sensitive data in mind.

Analyze a file in a few steps

Open the conversation you want to work in.
Attach your file. In the message box, select the + menu and choose Attach file — or drag the file straight onto the message box (you can also paste it from your clipboard). A chip appears showing its progress: uploading → scanning → ready.
Wait for the chip to read ready. Every upload is scanned for safety before it’s available; this usually takes a few seconds.
Ask a specific analysis question. Name the columns and the calculation you want. For example: “Using the attached sales export, total revenue by region and show the result as a table, sorted highest to lowest.”
Send, and read the answer. Chat returns the figures — usually as a table — and explains what it found.
Keep going. Follow-up questions reuse the same file, so you don’t re-attach it. Drill in: “Now break the top region down by product,” or “Which months were below the yearly average?”

Tip — point to the columns. Chat answers best when you use the actual column names from your file (“group by Region, sum Net Amount”). Vague asks like “analyze this” give vague results.

Ask for the analysis you actually want

The clearer your request, the more precise the result. Some patterns that work well:

You want	Ask something like
A summary of the whole sheet	”Describe this dataset: how many rows, what each column means, and anything that looks off.”
A total or average	”What’s the total `Amount`, and the average per `Customer`?”
A breakdown	”Sum `Revenue` by `Region`, as a table sorted from highest to lowest.”
A trend over time	”Show monthly `Signups` for this year and tell me which months grew and which shrank.”
Outliers or errors	”List any rows where `Quantity` is negative or `Email` is blank.”
A comparison	”Compare `Q1` to `Q2` for each product and show the percentage change.”
A ranked list	”Top 10 customers by total spend, with their order count.”

You can also ask Chat to show its working — “show the rows that make up that total” — so you can verify the figures yourself.

Get tables in responses

When an answer is naturally tabular — a breakdown, a ranking, a comparison — Chat formats it as a real table right in the conversation, with aligned columns and headers. The easiest way to get one is simply to ask:

“Show this as a table with one row per product.”
“Break revenue down by month and put it in a table, highest first.”
“Compare Q1 and Q2 side by side.”

Once you have a table, you can:

Copy the response with the copy button beneath the message, then paste it into a spreadsheet, document, or email — the table structure comes across cleanly.
Export the whole conversation to PDF, Markdown, or text when you want to hand the analysis off — see Export a conversation.

If a table comes back wider or longer than you need, just ask: “Show only the top 5 rows,” or “drop the ID column.”

A worked example

You’ve been handed q2-orders.xlsx with columns Date, Region, Product, and Amount, and you need a regional summary for a meeting.

Open a new chat, drag q2-orders.xlsx onto the message box, and wait for the chip to read ready.
Ask:

Summarize total Amount by Region as a table, sorted highest to lowest, and add a grand-total row. Then tell me which region grew most versus Q1.
Chat returns a clean table — each region with its total, plus the grand total — and a one-line read on the biggest mover.
Follow up without re-attaching:

Now break the top region down by Product.
Copy the table into your deck, or export the conversation to keep the full working.

Go deeper when you need to

For richer questions — multi-step reasoning, cross-checking, or pulling in outside context alongside your data — choose a higher interaction tier before you send, using the tier selector at the bottom-right of the message box:

Interactive Mode — fastest, best for quick lookups.
Thinking — the default; good for most analysis with several steps.
Research — adds web search when your question needs outside facts as well as your data.

To learn what each tier does, see The surfaces.

Reuse a data file across conversations

If it’s a dataset you’ll come back to — a reference list, a recurring export — you don’t have to re-attach it every time. Promote it into a Knowledge collection once, and Chat can draw on it across conversations and cite it back to you. See Upload a file to work with for how attachments and Knowledge differ, and how to promote one.

You can also save a reusable prompt like “Summarize Amount by Region as a sorted table” so you can run the identical analysis on next period’s file — see Saved prompts.

Supported data formats

Category	Formats
Spreadsheets	XLS, XLSX, CSV
Documents with tables	PDF, DOC, DOCX, PPT, PPTX, TXT
Structured data	JSON, XML, YAML / YML, Markdown (MD), HTML

Size limit: 100 MB per file. Excel and other Office files are converted to text behind the scenes so Chat can read and search them. For the cleanest results, keep one table per sheet, put column names in the first row, and avoid merged cells or blank header rows.

Keep sensitive data in mind

A one-off attachment stays within that conversation and the people in it; a Personal Knowledge collection is private to you alone — even organization admins don’t get access.
Promoting data to Knowledge widens who can see it (Folder scope shares it with a project team; Tenant scope with your whole organization). Promotion is one-way, so choose the scope deliberately.
If your organization has a policy on what may be uploaded to AI tools, follow it. When in doubt, start narrow — see What not to share.

Troubleshooting

What you see	What it means	What to do
The answer ignores some columns	The question was too broad, or the columns weren’t named	Reference the exact column names and the calculation you want
Numbers look off	Mixed formats, hidden blanks, or stray text in numeric columns; or a messy layout	Ask Chat to “show the rows behind that total”; put one clean table per sheet with a single header row, then re-attach
”Answering from general knowledge” notice	The file may still be processing, or wasn’t picked up	Wait for the chip to read ready, then re-ask and point to the file by name
The table is too wide to read	The result has more columns than you need	Ask for a subset: “show only `Region` and `Total`”
Chip stuck on scanning, or shows timed out	The safety scan is running or didn’t finish in time	Give a large file a moment; if it times out, the file is usually still sendable — if results look off, remove it and try again
Error: not a supported file type	The format isn’t on the supported list	Export it to CSV or XLSX and try again
Error: exceeds the 100 MB limit	The file is too big	Split it into smaller files, or filter it down before uploading

Good to know: Your uploaded data stays inside your organization’s workspace and is handled under the same protections as the rest of your conversations. To see what that covers, read Your privacy.

Upload a file to work with — the basics of attaching files and promoting them to Knowledge
Summarize a document — fast, grounded summaries of longer files
Export a conversation — hand off your analysis as PDF, Markdown, or text
Saved prompts — re-run the same analysis on next period’s data